4 ms·

FB just recently shot their credibility though by making statements that seemed very dishonest regarding their content bias. I don't think it's a big issue but

FB just recently shot their credibility though by making statements that seemed very dishonest regarding their content bias. I don't think it's a big issue but it'll look that way. It'll cause people to think twice -- who knows, maybe next week FB will announce "a bug in the code sometimes accidentally listened to a few users' phone calls" and "we are reviewing our auditing mechanisms to ensure this doesn't happen again". Is it a net gain for society if they're more wary, even if this particular instance is wrong?

abalone 10y ago
What were the "very dishonest" statements?
- MichaelGG 10y ago
  Saying that there was no bias in their news selection, when it should be readily apparent that having humans select thing would introduce bias. Not quite outright lying, as you might possibly think they're naive enough to think humans are unbiased, but I don't think they get the benefit of the doubt here. IIRC they also started off saying it was all algorithmic.
  - abalone 10y ago
    Can you cite where they said it was "all" algorithmic? Their investigation found "no evidence of systematic political bias." So, not sure where the very dishonest statement is here.
    - _yvjs 10y ago
      The point he's making is that Facebook can't be trusted to self-regulate (in the context of privacy). With that assumption you probably wouldn't accept their opaque self-investigation so placidly. I think the point is that a casual observer would not have guessed trending tropics were so influenced by human intervention, and might have taken them to be an actual reflection of the Facebook hivemind. Further the statement of "no evidence of systematic political bias" is a bit hard to swallow given reports of Facebook employees asking "what Facebook can do to stop Donald Trump" at their all-hands (one of the top 5 questions in a poll [0] not just a lone employee walking up to the microphone). Further, the people curating these feeds are probably underpaid and young, and hence are probably going to skew left of the average journalist. Bias doesn't have to be explicit to be systematic. Anyway I find Donald Trump + much of the rest of the GOP field pretty reprehensible politics-wise, Facebook seems to be handling things well [1], and we shouldn't worry about a conflict of interest as much as w/ advertising+privacy [2], but some amount of vigilance seems wise given how much influence these platforms can have over political outcomes [3]. Twitter in particular seems to have made a few fairly illiberal moves lately from my POV. [0] http://gizmodo.com/facebook-employees-asked-mark-zuckerberg-if-they-should-1771012990 http://gizmodo.com/facebook-employees-asked-mark-zuckerberg-... [1] https://medium.com/@glennbeck/what-disturbed-me-about-the-facebook-meeting-3bbe0b96b87f https://medium.com/@glennbeck/what-disturbed-me-about-the-fa... [2] One shouldn't discount the H1B angle completely though. [3] http://www.politico.com/magazine/story/2015/08/how-google-could-rig-the-2016-election-121548 http://www.politico.com/magazine/story/2015/08/how-google-co...
- reitanqild 10y ago
  For a very real example of facebook bias a group created two Facebook groups full of similar hatred but directed at two different ethnic groups. One was shut down almost immediately, one was still up when the video was posted. Won't name them as I don't want to drag politics into HN but I am fairly convinced it is repeatable (and if anyone is interested in giving it a try I can try to dig out a reference and post.)
  - darkandbrooding 10y ago
    Interesting experiment, but two groups seems like a too small a sample. If moderation is done by humans, you could simply prove that one moderator had a worse morning than the other, or that through luck of the draw Group A went to a sympathetic moderator while group B did not. That could demonstrate individual bias, but not institutional. If moderation is automated, how do you control for the weight of keywords, phrases, etc? Maybe Group A was full of statements that almost threw a flag but not quite, while Group B contained mostly innocuous wording but (only) one phrase that actually got flagged. If both groups were designed to be unpleasant, but various "blacklist" words and phrases are weighted differently, Group A might have achieved "flagged" weight while group Group B was "flagged -1." One might have scored 99 pts vs 100 for the other, but the visible difference is that one remains visible while the other does not. Thinking out loud, now... let's say that Facebook does have a political bias. How do you prove that Facebook's financial success is despite that political bias, and not because of it? Is it possible to have a monopoly on digital socialization? If you can't declare Facebook a monopoly, and therefore subject to greater regulation, then your only lever over Facebook is social pressure, presumably in the form of bad press and boycotts, which will reduce their advertising revenue. How do you prove that the existing political bias is leading them to sub-optimal revenues?
    - reitanqild 10y ago
      > Maybe Group A was full of statements that almost threw a flag but not quite, while Group B contained mostly innocuous wording but (only) one phrase that actually got flagged. Couple of months since I saw this, so can't say for sure but I think a major point of the experiment was that the two "hate groups" were equal except which ethnic group they were directed at.
tremon 10y ago
maybe next week FB will announce "a bug in the code" Or they will explain how it was "a rogue employee" who put the code there undetected, and caused it to spread to millions of devices.

abalone 10y ago

What were the "very dishonest" statements?

MichaelGG 10y ago
Saying that there was no bias in their news selection, when it should be readily apparent that having humans select thing would introduce bias. Not quite outright lying, as you might possibly think they're naive enough to think humans are unbiased, but I don't think they get the benefit of the doubt here. IIRC they also started off saying it was all algorithmic.
- abalone 10y ago
  Can you cite where they said it was "all" algorithmic? Their investigation found "no evidence of systematic political bias." So, not sure where the very dishonest statement is here.
  - _yvjs 10y ago
    The point he's making is that Facebook can't be trusted to self-regulate (in the context of privacy). With that assumption you probably wouldn't accept their opaque self-investigation so placidly. I think the point is that a casual observer would not have guessed trending tropics were so influenced by human intervention, and might have taken them to be an actual reflection of the Facebook hivemind. Further the statement of "no evidence of systematic political bias" is a bit hard to swallow given reports of Facebook employees asking "what Facebook can do to stop Donald Trump" at their all-hands (one of the top 5 questions in a poll [0] not just a lone employee walking up to the microphone). Further, the people curating these feeds are probably underpaid and young, and hence are probably going to skew left of the average journalist. Bias doesn't have to be explicit to be systematic. Anyway I find Donald Trump + much of the rest of the GOP field pretty reprehensible politics-wise, Facebook seems to be handling things well [1], and we shouldn't worry about a conflict of interest as much as w/ advertising+privacy [2], but some amount of vigilance seems wise given how much influence these platforms can have over political outcomes [3]. Twitter in particular seems to have made a few fairly illiberal moves lately from my POV. [0] http://gizmodo.com/facebook-employees-asked-mark-zuckerberg-if-they-should-1771012990 http://gizmodo.com/facebook-employees-asked-mark-zuckerberg-... [1] https://medium.com/@glennbeck/what-disturbed-me-about-the-facebook-meeting-3bbe0b96b87f https://medium.com/@glennbeck/what-disturbed-me-about-the-fa... [2] One shouldn't discount the H1B angle completely though. [3] http://www.politico.com/magazine/story/2015/08/how-google-could-rig-the-2016-election-121548 http://www.politico.com/magazine/story/2015/08/how-google-co...
reitanqild 10y ago
For a very real example of facebook bias a group created two Facebook groups full of similar hatred but directed at two different ethnic groups. One was shut down almost immediately, one was still up when the video was posted. Won't name them as I don't want to drag politics into HN but I am fairly convinced it is repeatable (and if anyone is interested in giving it a try I can try to dig out a reference and post.)
- darkandbrooding 10y ago
  Interesting experiment, but two groups seems like a too small a sample. If moderation is done by humans, you could simply prove that one moderator had a worse morning than the other, or that through luck of the draw Group A went to a sympathetic moderator while group B did not. That could demonstrate individual bias, but not institutional. If moderation is automated, how do you control for the weight of keywords, phrases, etc? Maybe Group A was full of statements that almost threw a flag but not quite, while Group B contained mostly innocuous wording but (only) one phrase that actually got flagged. If both groups were designed to be unpleasant, but various "blacklist" words and phrases are weighted differently, Group A might have achieved "flagged" weight while group Group B was "flagged -1." One might have scored 99 pts vs 100 for the other, but the visible difference is that one remains visible while the other does not. Thinking out loud, now... let's say that Facebook does have a political bias. How do you prove that Facebook's financial success is despite that political bias, and not because of it? Is it possible to have a monopoly on digital socialization? If you can't declare Facebook a monopoly, and therefore subject to greater regulation, then your only lever over Facebook is social pressure, presumably in the form of bad press and boycotts, which will reduce their advertising revenue. How do you prove that the existing political bias is leading them to sub-optimal revenues?
  - reitanqild 10y ago
    > Maybe Group A was full of statements that almost threw a flag but not quite, while Group B contained mostly innocuous wording but (only) one phrase that actually got flagged. Couple of months since I saw this, so can't say for sure but I think a major point of the experiment was that the two "hate groups" were equal except which ethnic group they were directed at.

tremon 10y ago

maybe next week FB will announce "a bug in the code" Or they will explain how it was "a rogue employee" who put the code there undetected, and caused it to spread to millions of devices.